AIモデルのベンチマーク

59 モデル · 14 ベンチマーク · 取得日

Terminal-Bench 4.0

ターミナルを使った複数段階の実務タスク。

出典: Vals AI ↗ · 評価版 4.0 · 出典更新

66タスクを mini-swe-agent で実行。pass@1 を3回測定した平均(avg@3)。

26 / 26 件

費用とスコアの比較

25 件 · 費用未公表 1 件

0255075100$0.00$7.9643$15.9285$23.8928$31.857スコア(%) ↑評価の実行費用(USD / 問) →Kimi K3.1 Pro · Effort max · 63.64% · $4.57Gemini 4 Carbon · Effort high · 63.13% · $7.45GPT-6.1 Sol · Effort max · 55.05% · $1.72GPT-6.1 Terra · Effort max · 49.49% · $2.16GPT-6 Sol · Effort max · 44.44% · $5.79GPT-5.6 Sol · Effort max · 37.88% · $7.98GLM 5.3 Flash · Effort max · 37.37% · $1.22Claude Haiku 5.5 · Effort max · 36.36% · $3.12DeepSeek V4.1 Flash 1017 · Effort max · 31.82% · $0.54MiMo V2.6 Pro · Effort 未公表 · 31.31% · $0.46MiniMax-M3.1-Flash · Effort max · 25.25% · $0.73MiMo V2.6 Flash · Effort 未公表 · 24.24% · $0.21DeepSeek V4.1 Flash · Effort high · 19.70% · $0.50Claude Opus 5.5 · Effort max · 65.15% · $13.20Claude Sonnet 5.5 · Effort max · 64.14% · $16.51GPT-6 Astra · Effort max · 59.60% · $9.58Wonder Full · Effort max · 58.08% · $10.53Claude Fable 5.1 · Effort max · 58.08% · $17.18Gemini 4 Argon · Effort high · 57.58% · $17.64Claude Opus 5 · Effort max · 53.53% · $18.60Wonder · Effort max · 49.49% · $14.66Claude Fable 5 · Effort max · 41.41% · $30.34GLM 5.3 · Effort max · 38.89% · $9.37Qwen 3.8 Max 0902 · Effort 未公表 · 38.89% · $12.15Qwen 3.8 Max · Effort 未公表 · 34.34% · $10.63
Claude Opus 5.5AnthropicEffort: max

65.15% / $13.20

  • OpenAI
  • Anthropic
  • Google
  • DeepSeek
  • その他

モデル別の結果

Terminal-Bench 4.0。出典Vals AI、更新日 2026-10-29。
モデル / 評価設定Effort
推論強度
スコア(%)実行費用
USD / 問
ハーネス
Claude Opus 5.5 ↗

Anthropic

max65.15%$13.20Mini-SWE-agent
Claude Sonnet 5.5 ↗

Anthropic

max64.14%$16.51Mini-SWE-agent
Kimi K3.1 Pro ↗

Moonshot AI

max63.64%$4.57Mini-SWE-agent
Gemini 4 Carbon ↗

Google

high63.13%$7.45Mini-SWE-agent
GPT-6 Astra ↗

OpenAI

max59.60%$9.58Mini-SWE-agent
Claude Fable 5.1 ↗

Anthropic

max58.08%$17.18Mini-SWE-agent
Wonder Full ↗

Meta

max58.08%$10.53Mini-SWE-agent
Gemini 4 Argon ↗

Google

high57.58%$17.64Mini-SWE-agent
GPT-6.1 Sol ↗

OpenAI

max55.05%$1.72Mini-SWE-agent
Claude Opus 5 ↗

Anthropic

max53.53%$18.60Mini-SWE-agent
GPT-6.1 Terra ↗

OpenAI

max49.49%$2.16Mini-SWE-agent
Wonder ↗

Meta

max49.49%$14.66Mini-SWE-agent
GPT-6 Sol ↗

OpenAI

max44.44%$5.79Mini-SWE-agent
Claude Fable 5 ↗

Anthropic

max41.41%$30.34Mini-SWE-agent
GLM 5.3 ↗

Z.ai

max38.89%$9.37Mini-SWE-agent
Qwen 3.8 Max 0902 ↗

Alibaba

未公表38.89%$12.15Mini-SWE-agent
GPT-5.6 Sol ↗

OpenAI

max37.88%$7.98Mini-SWE-agent
GLM 5.3 Flash ↗

Z.ai

max37.37%$1.22Mini-SWE-agent
Claude Haiku 5.5 ↗

Anthropic

max36.36%$3.12Mini-SWE-agent
Qwen 3.8 Max ↗

Alibaba

未公表34.34%$10.63Mini-SWE-agent
DeepSeek V4.1 Flash 1017 ↗

DeepSeek

max31.82%$0.54Mini-SWE-agent
MiMo V2.6 Pro ↗

Xiaomi

未公表31.31%$0.46Mini-SWE-agent
MiniMax-M3.1-Flash ↗

MiniMax

max25.25%$0.73Mini-SWE-agent
MiMo V2.6 Flash ↗

Xiaomi

未公表24.24%$0.21Mini-SWE-agent
DeepSeek V4.1 Flash ↗

DeepSeek

high19.70%$0.50Mini-SWE-agent
Beam 1.0 ↗

Reflection

未公表4.04%未公表Mini-SWE-agent
ベンチマーク横断のスコア一覧(59 モデル)
  • 1位
  • 2位
  • 3位
モデルごとのベンチマークスコアと Effort
モデルTerminal-Bench 4.0 ↗スコア(%)ProofBench v1.1 ↗スコア(%)Terminal-Bench-Science ↗スコア(%)Finance Agent v2.1 ↗スコア(%)GPQA Diamond ↗スコア(%)FrontierCode 2.0 ↗マージ適合スコア(%)AA-Briefcase v1.1 ↗AA-Briefcase EloAutomationBench-AA ↗目的達成スコア(%)HLE-Diamond ↗正答率(%)AA-Omniscience v1.1 ↗Net Score(−100〜100)GDP.pdf ↗All-pass(%)CritPt v2 Hard Subset ↗正答率(%)Vibe Code Bench 1-100 ↗連続完了スコア(%)HLAB-Vals-1.1 ↗タスク完全達成率(%)
Qwen 3.8 MaxAlibaba34.34%Effort: 未公表未掲載12.86%Effort: 未公表未掲載93.69%Effort: 未公表未掲載未掲載未掲載未掲載未掲載未掲載未掲載12.83%Effort: 未公表未掲載
Qwen 3.8 Max 0902Alibaba38.89%Effort: 未公表未掲載未掲載未掲載未掲載23.42%Effort: 未公表1608.08 EloEffort: 未公表56.19%Effort: 未公表24.00%Effort: 未公表10.47Effort: 未公表22.80%Effort: 未公表28.50%Effort: 未公表24.83%Effort: 未公表15.42%Effort: 未公表
Qwen3.8 27BAlibaba未掲載未掲載未掲載未掲載未掲載14.91%Effort: xhigh1393.08 EloEffort: xhigh48.24%Effort: xhigh15.30%Effort: xhigh-6.37Effort: xhigh16.40%Effort: xhigh4.00%Effort: xhigh未掲載未掲載
Claude Fable 5Anthropic41.41%Effort: max95.00%Effort: max15.71%Effort: max未掲載93.18%Effort: max未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載
Claude Fable 5.1Anthropic58.08%Effort: max1位100.00%Effort: max40.00%Effort: max未掲載93.43%Effort: max31.62%Effort: maxDefault Fallback1669.80 EloEffort: maxDefault Fallback59.38%Effort: maxDefault Fallback60.10%Effort: maxDefault Fallback3位46.27Effort: maxDefault Fallback26.20%Effort: maxDefault Fallback51.00%Effort: maxDefault Fallback28.00%Effort: max未掲載
Claude Haiku 5.5Anthropic36.36%Effort: max89.00%Effort: max10.00%Effort: max65.37%Effort: max未掲載34.93%Effort: max1575.90 EloEffort: max35.41%Effort: max29.60%Effort: max12.30Effort: max20.80%Effort: max41.00%Effort: max20.41%Effort: max4.17%Effort: max
Claude Opus 4.8Anthropic未掲載未掲載未掲載未掲載92.42%Effort: max未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載
Claude Opus 5Anthropic53.53%Effort: max99.00%Effort: max27.14%Effort: max未掲載93.43%Effort: 未公表未掲載未掲載未掲載未掲載未掲載未掲載未掲載3位28.53%Effort: max未掲載
Claude Opus 5.5Anthropic1位65.15%Effort: max1位100.00%Effort: max3位47.14%Effort: max1位67.85%Effort: max未掲載1位46.44%Effort: maxDefault Fallback2位1800.51 EloEffort: maxDefault Fallback69.54%Effort: maxDefault Fallback2位67.90%Effort: maxDefault Fallback2位48.13Effort: maxDefault Fallback26.20%Effort: maxDefault Fallback61.00%Effort: maxDefault Fallback2位30.36%Effort: xhigh9.17%Effort: max
Claude Sonnet 5Anthropic未掲載77.00%Effort: max未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載13.82%Effort: max未掲載
Claude Sonnet 5.5Anthropic2位64.14%Effort: max1位100.00%Effort: max3位47.14%Effort: max67.28%Effort: max未掲載41.70%Effort: maxDefault Fallback1位1813.57 EloEffort: maxDefault Fallback71.77%Effort: maxDefault Fallback53.30%Effort: maxDefault Fallback30.47Effort: maxDefault Fallback25.80%Effort: maxDefault Fallback56.00%Effort: maxDefault Fallback未掲載10.42%Effort: max
Nemotron 3.5 Super VLBaseten未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載3.75%Effort: high
DeepSeek V4 Pro 0813DeepSeek未掲載未掲載未掲載未掲載92.42%Effort: max未掲載未掲載未掲載未掲載未掲載未掲載未掲載17.53%Effort: max10.83%Effort: max
DeepSeek V4.1 FlashDeepSeek19.70%Effort: high54.00%Effort: high0.00%Effort: max65.43%Effort: max未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載16.38%Effort: high未掲載
DeepSeek V4.1 Flash 1017DeepSeek31.82%Effort: max未掲載未掲載62.93%Effort: max未掲載28.47%Effort: max1570.42 EloEffort: max66.64%Effort: max25.30%Effort: max0.90Effort: max17.20%Effort: max24.00%Effort: max未掲載14.17%Effort: max
Gemini 3 Pro (11/25)Google未掲載未掲載未掲載未掲載91.67%Effort: high未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載
Gemini 3.1 Pro Preview (02/26)Google未掲載未掲載未掲載未掲載1位95.45%Effort: high未掲載未掲載未掲載未掲載未掲載未掲載未掲載6.69%Effort: high未掲載
Gemini 3.5 FlashGoogle未掲載未掲載未掲載未掲載92.68%Effort: high未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載
Gemini 3.6 FlashGoogle未掲載未掲載未掲載未掲載93.43%Effort: high未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載
Gemini 3.7 FlashGoogle未掲載未掲載未掲載未掲載93.94%Effort: high未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載
Gemini 3.8 FlashGoogle未掲載未掲載8.57%Effort: high65.76%Effort: high94.44%Effort: high18.74%Effort: high1202.78 EloEffort: high59.93%Effort: high33.10%Effort: high23.27Effort: high21.00%Effort: high43.50%Effort: high18.77%Effort: high12.08%Effort: high
Gemini 4 ArgonGoogle57.58%Effort: high99.00%Effort: high44.29%Effort: high2位67.66%Effort: high未掲載未掲載未掲載2位77.51%Effort: high未掲載未掲載未掲載未掲載21.20%Effort: high1位27.08%Effort: high
Gemini 4 CarbonGoogle63.13%Effort: high未掲載未掲載67.16%Effort: high未掲載40.56%Effort: high1610.19 EloEffort: high1位77.82%Effort: high1位69.10%Effort: high1位53.50Effort: high28.40%Effort: high63.00%Effort: high未掲載17.50%Effort: high
Mercury 3Inception未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載5.00%Effort: high
Ling 3.1 Flash 1019InclusionAI未掲載未掲載未掲載未掲載未掲載17.01%Effort: 未公表1521.70 EloEffort: 未公表67.55%Effort: 未公表17.20%Effort: 未公表-0.13Effort: 未公表12.60%Effort: 未公表37.50%Effort: 未公表未掲載未掲載
K2 Horizon 375B A23BInstitute of Foundation Models未掲載未掲載未掲載未掲載未掲載11.85%Effort: 未公表1293.06 EloEffort: 未公表37.21%Effort: 未公表10.20%Effort: 未公表-5.00Effort: 未公表7.40%Effort: 未公表11.50%Effort: 未公表未掲載未掲載
Muse Spark 1.3 MaxMeta未掲載未掲載10.00%Effort: max未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載20.46%Effort: max未掲載
Muse Spark OpenMeta未掲載未掲載未掲載60.66%Effort: xhigh未掲載7.60%Effort: xhigh1279.13 EloEffort: xhigh49.71%Effort: xhigh26.30%Effort: xhigh25.80Effort: xhigh18.60%Effort: xhigh27.50%Effort: xhigh未掲載2位25.00%Effort: xhigh
WonderMeta49.49%Effort: max84.00%Effort: max22.86%Effort: max65.55%Effort: max未掲載27.88%Effort: max1658.65 EloEffort: max68.09%Effort: max39.00%Effort: max25.10Effort: max30.60%Effort: max48.00%Effort: max26.55%Effort: max3位23.75%Effort: max
Wonder FullMeta58.08%Effort: max94.00%Effort: max37.14%Effort: max3位67.58%Effort: max未掲載37.76%Effort: max3位1729.32 EloEffort: max71.33%Effort: max48.50%Effort: max30.60Effort: max1位32.60%Effort: max53.00%Effort: max1位33.98%Effort: max23.33%Effort: max
MiniMax-M3MiniMax未掲載未掲載未掲載未掲載92.68%Effort: 未公表未掲載未掲載未掲載未掲載未掲載未掲載未掲載9.22%Effort: 未公表未掲載
MiniMax-M3.1-FlashMiniMax25.25%Effort: max未掲載未掲載58.83%Effort: max未掲載20.82%Effort: 未公表1474.82 EloEffort: 未公表72.52%Effort: 未公表28.10%Effort: 未公表-6.40Effort: 未公表20.80%Effort: 未公表31.00%Effort: 未公表未掲載7.50%Effort: max
Mistral Large 4Mistral AI未掲載未掲載未掲載61.38%Effort: high未掲載14.96%Effort: high1349.63 EloEffort: high57.83%Effort: high17.30%Effort: high5.13Effort: high17.60%Effort: high26.00%Effort: high14.26%Effort: high17.50%Effort: high
Mistral Large 4 OSSMistral AI未掲載未掲載未掲載61.72%Effort: high未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載15.83%Effort: high
Kimi K3Moonshot AI未掲載87.00%Effort: max未掲載未掲載92.93%Effort: 未公表未掲載未掲載未掲載未掲載未掲載未掲載未掲載18.24%Effort: max12.92%Effort: max
Kimi K3.1 ProMoonshot AI3位63.64%Effort: max91.00%Effort: max31.43%Effort: max61.49%Effort: max未掲載25.45%Effort: 未公表1675.78 EloEffort: 未公表3位73.30%Effort: 未公表37.30%Effort: 未公表29.87Effort: 未公表2位31.40%Effort: 未公表45.00%Effort: 未公表未掲載11.67%Effort: max
Nemotron 3.5 Super VLNVIDIA未掲載未掲載未掲載未掲載未掲載6.11%Effort: 未公表998.76 EloEffort: 未公表14.82%Effort: 未公表8.80%Effort: 未公表-2.87Effort: 未公表8.00%Effort: 未公表13.50%Effort: 未公表未掲載未掲載
GPT 5.2OpenAI未掲載未掲載未掲載未掲載91.67%Effort: xhigh未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載
GPT 5.5OpenAI未掲載未掲載未掲載未掲載93.18%Effort: xhigh未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載
GPT-5.6 LunaOpenAI未掲載未掲載未掲載未掲載91.67%Effort: max未掲載未掲載未掲載未掲載未掲載未掲載未掲載22.59%Effort: max未掲載
GPT-5.6 SolOpenAI37.88%Effort: max83.00%Effort: max20.00%Effort: max未掲載2位95.20%Effort: max未掲載未掲載未掲載未掲載未掲載未掲載未掲載20.01%Effort: max未掲載
GPT-5.6 TerraOpenAI未掲載74.00%Effort: max10.00%Effort: max未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載14.82%Effort: max未掲載
GPT-6 AstraOpenAI59.60%Effort: max99.00%Effort: max1位62.86%Effort: max未掲載未掲載2位42.37%Effort: max1568.98 EloEffort: max68.49%Effort: max3位63.80%Effort: max42.43Effort: max3位31.00%Effort: max1位70.50%Effort: max27.64%Effort: max7.08%Effort: max
GPT-6 LunaOpenAI未掲載未掲載未掲載未掲載未掲載25.17%Effort: max1336.99 EloEffort: max53.17%Effort: max22.20%Effort: max-8.27Effort: max22.80%Effort: max41.50%Effort: max未掲載4.17%Effort: max
GPT-6 SolOpenAI44.44%Effort: max83.00%Effort: max30.00%Effort: max未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載
GPT-6.1 SolOpenAI55.05%Effort: max99.00%Effort: max2位52.86%Effort: max61.82%Effort: max未掲載40.63%Effort: max1555.26 EloEffort: max64.87%Effort: max58.80%Effort: max42.03Effort: max3位31.00%Effort: max2位68.50%Effort: max未掲載6.25%Effort: max
GPT-6.1 TerraOpenAI49.49%Effort: max未掲載未掲載未掲載未掲載3位42.32%Effort: max1402.26 EloEffort: max60.17%Effort: max47.30%Effort: max29.57Effort: max28.60%Effort: max3位64.50%Effort: max未掲載1.67%Effort: max
Beam 1.0Reflection4.04%Effort: 未公表18.00%Effort: 未公表1.43%Effort: 未公表51.83%Effort: 未公表未掲載5.53%Effort: 未公表976.20 EloEffort: 未公表48.46%Effort: 未公表12.80%Effort: 未公表2.33Effort: 未公表11.60%Effort: 未公表26.50%Effort: 未公表未掲載4.58%Effort: 未公表
Step 5 PreviewStepFun未掲載未掲載未掲載未掲載未掲載12.00%Effort: 未公表1424.55 EloEffort: 未公表50.96%Effort: 未公表24.30%Effort: 未公表16.43Effort: 未公表14.80%Effort: 未公表32.50%Effort: 未公表未掲載12.92%Effort: 未公表
Hy4 PreviewTencent未掲載75.00%Effort: 未公表未掲載61.62%Effort: 未公表未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載8.75%Effort: 未公表
InklingThinking Machines未掲載未掲載未掲載未掲載未掲載3.92%Effort: xhigh833.06 EloEffort: xhigh4.98%Effort: xhigh10.60%Effort: xhigh-1.50Effort: xhigh12.80%Effort: xhigh18.00%Effort: xhigh7.33%Effort: 0.993.75%Effort: 0.99
Inkling SmallThinkingmachines未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載2.92%Effort: 0.99
Grok 4.5xAI未掲載未掲載未掲載未掲載92.93%Effort: high未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載
Grok 4.6xAI未掲載未掲載11.43%Effort: high未掲載3位94.70%Effort: high未掲載未掲載未掲載未掲載未掲載未掲載未掲載14.75%Effort: high未掲載
Grok 4.7xAI未掲載未掲載10.00%Effort: xhigh未掲載未掲載28.66%Effort: xhigh1637.72 EloEffort: xhigh65.56%Effort: xhigh23.20%Effort: xhigh31.23Effort: xhigh20.00%Effort: xhigh32.00%Effort: xhigh未掲載18.33%Effort: xhigh
MiMo V2.6 FlashXiaomi24.24%Effort: 未公表63.00%Effort: 未公表4.29%Effort: 未公表63.32%Effort: 未公表未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載13.33%Effort: 未公表
MiMo V2.6 ProXiaomi31.31%Effort: 未公表70.00%Effort: 未公表2.86%Effort: 未公表64.26%Effort: 未公表未掲載23.97%Effort: 未公表1512.15 EloEffort: 未公表58.62%Effort: 未公表15.20%Effort: 未公表-4.47Effort: 未公表19.20%Effort: 未公表34.00%Effort: 未公表未掲載13.75%Effort: 未公表
GLM 5.3Z.ai38.89%Effort: max未掲載未掲載65.00%Effort: max未掲載21.59%Effort: max1504.57 EloEffort: max62.20%Effort: max17.90%Effort: max6.03Effort: max11.20%Effort: max33.50%Effort: max19.99%Effort: max14.58%Effort: max
GLM 5.3 FlashZ.ai37.37%Effort: max未掲載未掲載60.59%Effort: max未掲載未掲載未掲載未掲載未掲載未掲載未掲載未掲載16.03%Effort: max10.42%Effort: max

出典